CN115147809A

CN115147809A - Obstacle detection method, device, equipment and storage medium

Info

Publication number: CN115147809A
Application number: CN202210767833.5A
Authority: CN
Inventors: 赵丛杨; 王学辉; 李晓晖; 张晔; 王昊
Original assignee: Apollo Intelligent Technology Beijing Co Ltd
Current assignee: Apollo Intelligent Technology Beijing Co Ltd
Priority date: 2022-06-30
Filing date: 2022-06-30
Publication date: 2022-10-04
Anticipated expiration: 2042-06-30
Also published as: CN115147809B

Abstract

The disclosure provides an obstacle detection method, an obstacle detection device, obstacle detection equipment and a storage medium, and relates to the technical field of image processing, in particular to the fields of automatic driving, automatic parking and the like. The specific implementation scheme is as follows: acquiring an image around a vehicle body; determining an obstacle grounding point in an image around a vehicle body; coordinate transformation is carried out on grounding points of the obstacles in the image coordinate system, and the positions of the grounding points of the obstacles in the grid map coordinate system are obtained; determining the occupation probability of each pixel in a grid map based on the position of the grounding point of each obstacle, wherein the occupation probability represents the probability that the pixel is occupied by the obstacle, and one grid in the grid map corresponds to one pixel; and clustering the plurality of pixels based on the occupation probability of each pixel in the grid map to obtain an obstacle detection result. The present disclosure enables detection of obstacles around a vehicle body.

Description

Obstacle detection method, device, equipment and storage medium

Technical Field

The present disclosure relates to the field of image processing technologies, and in particular, to the fields of automatic driving, automatic parking, and the like.

Background

Obstacle detection is a key component for realizing vehicle use, and for example, obstacle detection is important in the processes of vehicle running, vehicle parking and the like, and can affect the safety and reliability of vehicle running and vehicle parking.

Disclosure of Invention

The present disclosure provides an obstacle detection method, apparatus, device, and storage medium.

According to a first aspect of the present disclosure, there is provided an obstacle detection method including:

acquiring an image around a vehicle body;

determining an obstacle grounding point in the image around the vehicle body;

coordinate transformation is carried out on grounding points of the obstacles in the image coordinate system, and the positions of the grounding points of the obstacles in the grid map coordinate system are obtained;

determining an occupation probability of each pixel in a grid map based on the position of a grounding point of each obstacle, wherein the occupation probability represents the probability that the pixel is occupied by the obstacle, and one grid in the grid map corresponds to one pixel;

and clustering the plurality of pixels based on the occupation probability of each pixel in the grid map to obtain an obstacle detection result.

According to a second aspect of the present disclosure, there is provided an obstacle detection device including:

the acquisition module is used for acquiring images around the vehicle body;

a first determination module for determining an obstacle grounding point in the image around the vehicle body;

the conversion module is used for carrying out coordinate conversion on the grounding points of the obstacles in the image coordinate system to obtain the positions of the grounding points of the obstacles in the grid map coordinate system;

a second determining module, configured to determine, based on a position of a grounding point of each obstacle, an occupation probability of each pixel in a grid map, where the occupation probability represents a probability that the pixel is occupied by the obstacle, and one grid in the grid map corresponds to one pixel;

and the clustering module is used for clustering a plurality of pixels based on the occupation probability of each pixel in the grid map to obtain an obstacle detection result.

According to a third aspect of the present disclosure, there is provided an electronic device comprising:

at least one processor; and

a memory communicatively coupled to the at least one processor; wherein,

the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method of the first aspect.

According to a fourth aspect of the present disclosure, there is provided a non-transitory computer readable storage medium having stored thereon computer instructions for causing the computer to perform the method of the first aspect.

According to a fifth aspect of the present disclosure, there is provided a computer program product comprising a computer program which, when executed by a processor, implements the method of the first aspect.

According to a sixth aspect of the present disclosure, there is provided a vehicle including: the electronic device of the third aspect.

According to a seventh aspect of the present disclosure, there is provided a cloud control platform comprising the electronic device according to the third aspect.

The present disclosure enables detection of obstacles around a vehicle body.

It should be understood that the statements in this section are not intended to identify key or critical features of the embodiments of the present disclosure, nor are they intended to limit the scope of the present disclosure. Other features of the present disclosure will become apparent from the following description.

Drawings

The drawings are included to provide a better understanding of the present solution and are not to be construed as limiting the present disclosure. Wherein:

fig. 1 is a flowchart of an obstacle detection method provided by an embodiment of the present disclosure;

FIG. 2 is a schematic diagram of detecting an image around a vehicle body through a detection network according to an embodiment of the disclosure;

FIG. 3 is a schematic diagram of relationships between different coordinate systems in an embodiment of the present disclosure;

FIG. 4 is a schematic illustration of determining an occupancy probability in an embodiment of the disclosure;

FIG. 5 is a schematic diagram of a grid map corresponding to different times in an embodiment of the present disclosure;

FIG. 6 is a schematic illustration of fusion for a grid map in an embodiment of the present disclosure;

fig. 7 is a schematic diagram of an obstacle detection method to which an embodiment of the present disclosure is applied;

fig. 8 is a schematic diagram of outputting obstacle detection results in an embodiment of the present disclosure;

fig. 9 is a schematic structural diagram of an obstacle detection device provided in an embodiment of the present disclosure;

fig. 10 is a block diagram of an electronic device for implementing the obstacle detection method of the embodiment of the present disclosure.

Detailed Description

Exemplary embodiments of the present disclosure are described below with reference to the accompanying drawings, in which various details of the embodiments of the disclosure are included to assist understanding, and which are to be considered as merely exemplary. Accordingly, those of ordinary skill in the art will recognize that various changes and modifications of the embodiments described herein can be made without departing from the scope and spirit of the disclosure. Also, descriptions of well-known functions and constructions are omitted in the following description for clarity and conciseness.

The disclosed embodiment provides an obstacle detection method, which may include:

acquiring an image around a vehicle body;

determining an obstacle grounding point in an image around a vehicle body;

determining the occupation probability of each pixel in a grid map based on the position of the grounding point of each obstacle, wherein the occupation probability represents the probability that the pixel is occupied by the obstacle, and one grid in the grid map corresponds to one pixel;

The embodiment of the disclosure realizes the detection of the obstacles around the vehicle body, and can provide guarantee for safe driving such as safe automatic driving.

Fig. 1 is a flowchart of an obstacle detection method provided in an embodiment of the present disclosure, and referring to fig. 1, the obstacle detection method provided in an embodiment of the present disclosure may include:

s101, acquiring an image around the vehicle body.

The image of the surroundings of the vehicle, i.e., the image of the surroundings of the vehicle, may include information on the surroundings of the vehicle.

The image around the vehicle body may be acquired by a laser radar, a camera, or the like.

One or more images of the surroundings of the vehicle body may be acquired.

And S102, determining the grounding point of the obstacle in the image around the vehicle body.

The ground point of an obstacle, i.e. the ground point of an obstacle, is understood to be the contact point of the obstacle with the ground.

In one implementation, S102 may include:

inputting the image around the vehicle body into a preset detection network, and outputting the grounding point scores of all points in the image around the vehicle body through the preset detection network, wherein the grounding point scores represent points which are the scores of the grounding points of the obstacles; determining an obstacle grounding point in the vehicle body surroundings image based on the grounding point fraction.

Determining the obstacle grounding point in the image around the vehicle body based on the grounding point fraction may include: and selecting a point with the grounding point fraction reaching a preset fraction as the grounding point of the obstacle. Or selecting preset points from each row of images around the vehicle body to form the grounding point of the obstacle.

For example, for each column, sorting the grounding point fractions of the midpoint of the column from high to low or from low to high, and if the sorting is performed in the order from high to low, selecting the point corresponding to the preset grounding point fraction sorted at the front; and if the ground points are sorted from low to high, selecting points corresponding to the sorted preset ground point fractions, and combining the points selected from each row to obtain the ground points of the obstacle. The number of the selected points in each column, that is, the number of the preset points, may be determined according to actual requirements or experience. In one example, for each column, the point with the highest fraction of ground points is selected from the column.

The predetermined detection network may be a pre-trained detection network. Specifically, a plurality of sample body peripheral images can be obtained, each sample body peripheral image is labeled, and a grounding point fraction corresponding to the sample body peripheral image, that is, a grounding point fraction true value corresponding to the sample body peripheral image is marked. And then, taking a sample vehicle body periphery image and the ground point fraction true value corresponding to the sample vehicle body periphery image as a sample pair, and training the neural network model based on a plurality of sample pairs to obtain the detection network. Specifically, the sample pair is input into the neural network model to obtain an output of the neural network model, the output is compared with a ground point score true value corresponding to the sample pair, for example, a difference between the two is calculated, and a model parameter is adjusted based on the difference, when a training end condition is met, a trained detection network is obtained, wherein the training end condition may include that the training time reaches a preset training time or that the difference obtained based on the sample pair is smaller than a preset threshold value.

The detection network can conveniently and accurately determine the grounding point fractions of all points in the images around the vehicle body, so that the determination of the grounding point of the obstacle in the images around the vehicle body based on the grounding point fractions is more convenient, and the determined grounding point of the obstacle is more accurate.

In an implementation manner, the network structure of the detection network may include a backbone network and a head network, where the backbone network is also called a backbone network and is mainly used for feature extraction to extract target features of different scales, different receptive fields, and different categories, so as to meet target detection requirements. The detection head is mainly used for predicting the result of the target, such as the grounding point fraction of the center point of the image around the vehicle body. In one example, the backbone uses the resource 50, feature pyramids are obtained after the backbone, each layer of the feature pyramids is up-sampled (resize), and the feature map size with the largest size (the lowest layer of the pyramid) is kept consistent. Then, the tensor is spliced, and a decode _ conv (the size of the convolution kernel is 1*1 or 3*3 is optional) is performed, wherein the size of the convolution kernel can be selected according to actual requirements, such as 1*1 or 3*3, and the result after convolution is subjected to head to predict the result.

As shown in fig. 2, in the embodiment of the present disclosure, the detection network may predict the grounding point fraction (pts _ source) of the midpoint in the image around the vehicle body, and may also predict other information, specifically, the detection network includes five heads, and the result after convolution respectively passes through the five heads, and a corresponding result is predicted. For example, a ground point fraction (pts _ source), a deviation of obstacle ground points due to scaling (pts _ bias), a category of obstacle ground points (pts _ class), an x-direction deviation of obstacle ground points from a frame center (pts _ offset _ x), and a y-direction deviation of obstacle ground points from a frame center (pts _ offset _ y) are predicted. Wherein the type of the obstacle grounding point represents the type of the obstacle, and the frame center represents the center of the obstacle detection frame. In addition, the results obtained by the five heads may be combined to output a VIS (Visual Identity System) recognition result.

And S103, performing coordinate transformation on the grounding points of the obstacles in the image coordinate system to obtain the positions of the grounding points of the obstacles in the grid map coordinate system.

And converting the coordinates of the grounding points of the obstacles in the image coordinate system based on the conversion relationship among the coordinate systems to obtain the positions of the grounding points of the obstacles in the grid map coordinate system.

In an alternative embodiment, S103 may include:

projecting the grounding points of the obstacles in the camera coordinate system to a spherical coordinate system to obtain the grounding points of the obstacles in the spherical coordinate system; calculating the coordinates of the grounding points of the obstacles in the spherical coordinate system in the ground coordinate system by utilizing the geometric similarity relation; and performing coordinate conversion on coordinates of the grounding points of the obstacles in the ground coordinate system to obtain the positions of the grounding points of the obstacles in the grid map coordinate system.

The simple understanding is that the pixel coordinates are firstly projected to the spherical surface to obtain the three-dimensional coordinates on the spherical surface, then the three-dimensional coordinates of the points on the ground are calculated by utilizing the geometric similarity relation such as triangle similarity, and then the coordinates under the ground coordinate system are converted into the grid map coordinate system.

Based on the relation between different coordinate systems, the position of the grounding point of the obstacle under the grid map can be conveniently obtained.

As shown in fig. 3, the coordinate system O-XYZ represents the coordinate system of the camera, the XOY plane of the Ground coordinate system coincides with the Ground plane, and the Z-axis is perpendicular to the Ground. Point P represents an obstacle on the ground, and may also be understood as an obstacle ground point. ouv represents the imaging plane, the camera will also invert the picture when saving the image, with the actual P' being the pixel location of the P point. The similarity relation of the triangles can be obtained through the geometric relation in the graph

Xp and Yp represent the coordinates of the P point in the ground coordinate system, and the coordinates can be solved through similarity of triangles. X _{imu_ground} 、Y _{imu_ground} 、Z _{imu_ground} 、O _{imu_ground} Representing the ground coordinate system, X _Q ,Y _Q 、Z _Q Representing the coordinates of the P point in a spherical coordinate system.

And S104, determining the occupation probability of each pixel in the grid map based on the position of the grounding point of each obstacle.

The occupancy probability represents the probability that a pixel is occupied by an obstacle.

One grid in the grid map corresponds to one pixel, and the occupation probability of each pixel in the grid map, namely the occupation probability of each grid is determined.

The position of the grounding point of each obstacle in the grid map coordinate system can also be understood as the pixel where the grounding point of each obstacle is located in the grid map.

In an alternative embodiment, S104 may include:

aiming at the position of a grounding point of each obstacle, connecting the origin and the position of a grid map coordinate system to obtain a grounding point connecting line; selecting a pixel area of which the angle between the grid map and a grounding point connecting line is smaller than a preset angle; determining pixels with the distance between the pixel area and the position smaller than a preset distance, and taking the pixels with the distance between the pixel area and the position smaller than the preset distance as pixels to be updated; the probability of occupation of the pixel to be updated is increased.

Specifically, an initial occupation probability may be preset, and the occupation probability of each pixel in the grid map is initialized first to be the initial occupation probability; aiming at the position of a grounding point of an obstacle, after determining pixels with the distance between the pixels and the position smaller than a preset distance in a pixel area, on the basis of the initial occupation probability, increasing the occupation probability of the pixels with the distance between the pixels and the position smaller than the preset distance in the pixel area. For example, the preset probability is increased on the basis of the initial occupancy probability. The preset probability may be determined according to actual requirements or experience, such as 0.5, and increasing the occupancy probability may also be understood as updating the occupancy probability. In this way, the probability of occupation of each pixel in the grid map determined based on the position of the grounding point of the obstacle is obtained. On the basis of the above, the probability of occupation of each pixel in the grid map is determined in the same manner as the probability of occupation of each pixel in the grid map is determined based on the position of the grounding point of the obstacle for the positions of the grounding points of other obstacles (positions other than the position of the grounding point of the above-mentioned obstacle among the positions of the grounding points of all obstacles), that is, after the pixel of which the distance from the position in the corresponding pixel region is smaller than the preset distance is obtained based on the position of the grounding point of one obstacle, the probability of occupation of the pixel to be updated is increased based on the probability of occupation of each pixel in the grid map determined based on the position of the grounding point of the previous obstacle, and when the probability of occupation of each pixel in the grid map is increased based on the positions of all obstacles, the probability of occupation of each pixel in the final grid map is obtained.

A process of determining an occupation probability of each pixel in a grid map based on a position of an obstacle ground point is shown in fig. 4, an origin and a position of a grid map coordinate system are connected to obtain a ground point connecting line, such as a position P, and an OP is connected, then a pixel area in the grid map, in which an angle between the grid map and the ground point connecting line is smaller than a preset angle, is selected, for example, a pixel area, in which an angle between the grid map and the ground point connecting line and the OP is smaller than 1 degree, such as an area 401 is selected, and a pixel occupation probability that a distance between a pixel in the area 401 and a point P is smaller than a threshold value is increased by 0.5. Intuitively, the probability of a pixel within arc 402 increases by 0.5.

The occupation probability of each pixel in the grid map can be conveniently and accurately updated based on the position of the grounding point of each obstacle.

And S105, clustering the plurality of pixels based on the occupation probability of each pixel in the grid map to obtain an obstacle detection result.

The clustering mode is not limited in the embodiments of the present disclosure, and any mode capable of realizing pixel clustering is within the scope of the embodiments of the present disclosure.

For example, the pixel clustering may adopt super-pixel clustering, K-means clustering, BFS (Width First Search) algorithm, etc., wherein the K-means clustering refers to clustering with K points in space as the center.

The obstacle detection result may be a convex hull obtained by clustering.

After obtaining the obstacle detection result, the obstacle detection result may be output, for example, a convex hull obtained by clustering in a grid map.

The vehicle can be decided and controlled according to the obstacle detection result, such as running, parking and the like, for example, after the vehicle is automatically driven to obtain the obstacle detection result, the vehicle can be controlled to avoid the obstacle and continue to move ahead, and the running safety is improved.

In an alternative embodiment S105 may comprise:

for each pixel, traversing the neighborhood of the pixel in response to the fact that the occupation probability of the pixel is larger than a preset occupation probability threshold, and storing the neighborhood pixels of which the occupation probability is larger than the preset occupation probability threshold in the neighborhood into a cluster corresponding to the pixel; clustering is carried out on the basis of neighborhood pixels in the clusters to obtain a clustering result corresponding to the pixels; and combining the clustering results corresponding to the pixels to obtain an obstacle detection result.

The preset occupancy probability threshold may be determined according to actual requirements or experience, etc. The neighborhood pixels may include pixels within a preset range from the pixel.

For example, traversing each pixel in the grid map, if the "occupation probability" of the pixel exceeds a preset threshold, that is, a preset occupation probability threshold, maintaining a cluster, traversing the neighborhood of the pixel, storing a matched pixel (a pixel whose occupation probability in the neighborhood is greater than the preset occupation probability threshold) in the neighborhood, popping up the pixel every time a pixel in the cluster is accessed, ending the cycle of the pixel after the number of elements in the cluster is reduced to zero, aggregating the pixels whose thresholds exceed the set threshold, and executing the process for each pixel, so that a polygon can be generated, that is, the clustering result corresponding to each pixel can be combined to obtain the obstacle detection result.

Clustering can be achieved more easily through the clusters, clustering can be performed more conveniently, and the neighborhood of the pixel with the occupation probability larger than the preset occupation probability threshold value is clustered, so that the obtained obstacle detection result is more accurate.

In an optional embodiment, S101 may include:

acquiring a plurality of images around a vehicle body;

s105 may include:

aligning the grid maps respectively corresponding to a plurality of moments to obtain a plurality of aligned pixels; aiming at each aligned pixel, fusing the occupation probabilities of the aligned pixels in different grid maps to obtain the occupation probability after fusion; and clustering the aligned pixels based on the fused occupation probability to obtain an obstacle detection result.

The plurality of images around the vehicle body may include images around the vehicle body captured at a plurality of times, for example, one image around the vehicle body captured at each time is selected for the plurality of times to form a plurality of images around the vehicle body.

And the grid map corresponding to each moment, namely the grid map corresponding to the image around the vehicle body at the moment. For example, the image around the vehicle body is coordinate-converted to obtain an image in a grid map coordinate system, that is, a grid map corresponding to the image around the vehicle body.

The grid maps corresponding to the multiple times are aligned to obtain multiple aligned pixels, and the pixels aligned with the pixels can be sequentially searched from other frame grid maps for each pixel in the frame grid map by taking one frame grid map as a reference. Specifically, the pixels in the grid map corresponding to the next time at the current time may be determined according to the traveling information of the vehicle, such as the traveling speed, the direction, and the like.

With a two-frame grid map alignment is illustrated as an example: calculating the corresponding relation of pixels in the grid maps corresponding to the front and rear time sequences (such as adjacent time sequences), namely aligning the grid maps corresponding to the two time sequences respectively to obtain a plurality of aligned pixels. For example, pixels in the grid map coordinate system may be transformed to the world coordinate system to facilitate the calculations. The pixels in the grid map coordinate system can be converted to the world coordinate system for convenient calculation.

For each aligned pixel, the occupation probabilities of the aligned pixels in different grid maps are fused to obtain the fused occupation probability, which can be calculated by the following formula:

wherein,

to post-fusion occupancy probability, P _t-1 Probability of occupation of a pixel in the grid map at time t-1, and P _t-1 And the occupation probability of the pixels in the grid map at the aligned t moment.

After the fusion-based occupation probability is obtained, clustering a plurality of pixels based on the occupation probability of each pixel in the grid map, and in the process of obtaining an obstacle detection result, replacing the occupation probability of each pixel in the grid map based on clustering with the fusion-based occupation probability, specifically, for each aligned pixel, traversing the neighborhood of the aligned pixel in response to the fact that the occupation probability of the aligned pixel is greater than a preset occupation probability threshold, and storing the neighborhood pixel of which the occupation probability is greater than the preset occupation probability threshold in the neighborhood into the cluster corresponding to the aligned pixel; clustering is carried out on the basis of neighborhood pixels in the clusters to obtain clustering results corresponding to the aligned pixels; and combining the clustering results corresponding to the aligned pixels to obtain an obstacle detection result.

As shown in FIG. 5, different grid maps correspond to different times, and the grid map corresponding to time t is map _t The grid map corresponding to the time t-1 is map _t-1 O in FIG. 5 _t-1 And O _t The arrows in between indicate the movement of the vehicle. Fusing the occupation probabilities of the aligned pixels in different grid maps to obtain the fused occupation probability; based on the merged occupation probability, clustering is performed on the plurality of aligned pixels to obtain an obstacle detection result, and the obstacle detection result obtained by clustering is shown in fig. 6.

Combining the images around the vehicle body, aiming at the aligned pixels of the grid maps, fusing the occupation probabilities of the aligned pixels in different grid maps, clustering the aligned pixels based on the fused occupation probabilities to obtain an obstacle detection result, and improving the accuracy of the obtained obstacle detection result.

In an alternative embodiment, the image of the periphery of the vehicle body may be captured by a fisheye camera, and S101 includes acquiring the image of the periphery of the vehicle body captured by the fisheye camera.

Projecting the grounding points of the obstacles in the camera coordinate system to the spherical coordinate system to obtain the grounding points of the obstacles in the spherical coordinate system, and the method comprises the following steps:

and projecting the grounding points of the obstacles in the coordinate system of the fisheye camera to the spherical coordinate system to obtain the grounding points of the obstacles in the spherical coordinate system.

Specifically, acquiring a vehicle body surrounding image acquired by a fisheye camera, determining grounding points of obstacles in the vehicle body surrounding image, projecting the grounding points of the obstacles under a fisheye camera coordinate system to a spherical coordinate system to obtain grounding points of the obstacles under the spherical coordinate system, and calculating coordinates of the grounding points of the obstacles under the spherical coordinate system under a ground coordinate system by using a geometric similarity relation; and performing coordinate conversion on coordinates of the grounding points of each obstacle in a ground coordinate system to obtain the position of the grounding point of each obstacle in a grid map coordinate system, determining the occupation probability of each pixel in the grid map based on the position of the grounding point of each obstacle, and determining the occupation probability of each pixel in the grid map based on the position of the grounding point of each obstacle.

It is simply understood that steps S102 to S105 are performed for the image around the vehicle body captured by the fisheye camera.

Gather the image around the automobile body through the fisheye camera to handle the image of gathering, with the barrier around the detection automobile body, because the field of view scope of fisheye camera is great, can gather to the barrier of laser radar, ordinary camera etc. the position that can not gather, so, can fill laser radar, ordinary camera and detect the blind area that exists to the barrier, provide the guarantee for safe autopilot.

In the related technology, the perception of the surrounding environment of the vehicle body can be realized by adopting a laser radar in the automatic driving process, and the three-dimensional information of obstacles around the vehicle body can be detected by the multi-line laser radar through point cloud data. However, because the laser radar has a blind area, the laser radar cannot realize better perception in an area within six meters around the vehicle body and a short obstacle nearby, so that the obstacle nearby the vehicle body cannot be detected, and the cost of the multi-line laser radar is high.

In the embodiment of the disclosure, the fisheye camera is used for detecting the grounding point of the near blind area obstacle, sensing the position of the blind area obstacle, providing guarantee for safe automatic driving, and in addition, reducing the cost of obstacle detection.

According to the embodiment of the invention, the images around the vehicle body collected by the fisheye camera are directly processed without firstly carrying out distortion removal on the images collected by the fisheye camera, so that the complexity of the calculation process is lower, and the calculation amount is smaller.

The automatic driving vehicle can effectively sense the environmental information around the vehicle body, accurately report the obstacle information around the automatic driving vehicle, and report the obstacle information by aiming at 360 degrees around the vehicle body of the fisheye camera, so that the problem of a blind area of a common sensor (a multi-line laser radar and a common camera) is effectively solved, and safety guarantee is provided for automatic driving.

One specific example is shown in fig. 7.

And acquiring images around the vehicle body through a fisheye camera.

And acquiring images around the vehicle body through the fisheye lens, such as fisheye forward view, fisheye left view, fisheye right view and fisheye back view.

The detection network comprises five heads, and the convolution results respectively pass through the five heads to predict corresponding results. For example, the deviation of the obstacle grounding points due to scaling (pts _ bias), the category of the obstacle grounding points (pts _ class), the x-direction deviation of the obstacle grounding points from the frame center (pts _ offset _ x), and the y-direction deviation of the obstacle grounding points from the frame center (pts _ offset _ y). The type of the grounding point of the obstacle represents the type of the obstacle, the center of the frame represents the center of the obstacle detection frame, and the results obtained by the five heads may be combined to output a VIS (Visual Identity System) recognition result.

Coordinate transformation is carried out on grounding points of the obstacles in the image coordinate system, and the positions of the grounding points of the obstacles in the grid map coordinate system are obtained; determining the occupation probability of each pixel in the grid map based on the position of the grounding point of each obstacle, wherein the occupation probability represents the probability that the pixel is occupied by the obstacle, and one grid in the grid map corresponds to one pixel; and clustering a plurality of pixels based on the occupation probability of each pixel in the grid map to obtain an obstacle detection result.

Specifically, the collected images around the car body are input into a pre-trained preset detection network to obtain detection results aiming at each point (including fraction of the grounding point, deviation of the grounding point caused by scaling, category of the grounding point, deviation of the grounding point in the x direction of the grounding point and the center of the frame, and deviation of the grounding point in the y direction of the center of the frame); determining a plurality of obstacle grounding points and clustering by using the detection result to obtain a plurality of target central points (central points of obstacle frames); coordinate transformation is carried out on grounding points of the obstacles in the image coordinate system, and the positions of the grounding points of the obstacles in the grid map coordinate system are obtained; determining a range of the occupation probability to be updated, namely determining a pixel to be updated, updating (such as increasing) the occupation probability of the pixel in the range, and clustering a plurality of pixels based on the occupation probability of each pixel in the grid map to obtain an obstacle detection result; or determining grid maps corresponding to a plurality of moments, aligning the grid maps, fusing the occupation probabilities of the aligned pixels, and clustering the aligned pixels by using the fused occupation probabilities to obtain an obstacle detection result. In this case, it is possible to specifically traverse each column of the tenter output of the first head, find the position of the largest pts _ source, which is considered as the grounding point of the obstacle, and then translate the position into the original image by scaling (scaling of the picture and deviation of the second head output). Since the feature map to size is 1/4 of the size of the input map, after conversion to the original map, there is not one obstacle ground point per column, but one obstacle ground point is reported every several columns. After the position is calculated, the type of the grounding point of the obstacle (the output of the third head) is also extracted according to the position information, and the grounding point of the obstacle to the type is obtained. And according to the output of the last two heads, aggregating the center of the target object (the center of the obstacle) by using a meanshift algorithm. And after the grounding point of the obstacle is obtained, taking out the offset from the output of the last two heads according to the position of the grounding point of the obstacle, obtaining a plurality of central points according to the offset, and then aggregating the points by using a meanshift algorithm to obtain the position of the final central point. The information of the center point position and the ground point of the obstacle can be transmitted to a downstream fusion module and a PNC (Planning and Control) module, and the automatic driving automobile can move according to the information.

The occupancy probability of each grid can be used for determining the position of the obstacle, namely, the plurality of pixels are clustered based on the occupancy probability of each pixel in the grid map, and the obstacle detection result is obtained. Other information such as category, sensor name, score and the like can also be transmitted to the fusion module together with the occupation probability and the obtained obstacle detection result to be fused with the point cloud information. The obstacle detection result may specifically be a clustered convex hull. As shown in fig. 8, the obstacle related information is output in the form of a 6-layer grid map, in which the aggregated convex hull is the obstacle detection result obtained above; the category represents the category of the obstacle, each point of the score is the score of the grounding point of the obstacle, and the category and the score can be obtained through the preset detection network; the sensor name may indicate the model of the fisheye camera, etc.

According to the embodiment of the disclosure, the sensing of the surrounding environment of the vehicle body is realized through the fisheye camera, the grounding point of the obstacle is detected based on the image around the vehicle body acquired by the fisheye camera, the coordinates of the grounding point of the obstacle under a ground coordinate system are calculated through a multi-view geometric method, and then the coordinates are converged into the convex hull to be reported, the field range of the fisheye camera is large, blind areas of the laser radar and the common camera can be filled, and for the obstacles (such as the conical roadblock) entering the blind areas, the related information of the obstacle can be reported, so that safety accidents are avoided.

Corresponding to the obstacle detection method provided by the above embodiment, an embodiment of the present disclosure further provides an obstacle detection apparatus, as shown in fig. 9, which may include:

an obtaining module 901, configured to obtain an image around a vehicle body;

a first determining module 902 for determining an obstacle grounding point in an image around a vehicle body;

a conversion module 903, configured to perform coordinate conversion on grounding points of the obstacles in the image coordinate system to obtain positions of the grounding points of the obstacles in the grid map coordinate system;

a second determining module 904, configured to determine, based on a position of a grounding point of each obstacle, an occupation probability of each pixel in a grid map, where the occupation probability represents a probability that a pixel is occupied by an obstacle, and one grid in the grid map corresponds to one pixel;

the clustering module 905 is configured to cluster the plurality of pixels based on the occupation probability of each pixel in the grid map to obtain an obstacle detection result.

Optionally, a second determination module 904, specifically for determining, for each obstacle ground point location, connecting the origin and the position of a grid map coordinate system to obtain a grounding point connection line; selecting a pixel area, of which the angle between the grid map and a grounding point connecting line is smaller than a preset angle; determining pixels with the distance between the pixel area and the position smaller than a preset distance, and taking the pixels with the distance between the pixel area and the position smaller than the preset distance as pixels to be updated; the probability of occupation of the pixel to be updated is increased.

Optionally, the clustering module 905 is specifically configured to, for each pixel, traverse a neighborhood of the pixel in response to that the occupation probability of the pixel is greater than a preset occupation probability threshold, and store a neighborhood pixel in the neighborhood whose occupation probability is greater than the preset occupation probability threshold into a cluster corresponding to the pixel; clustering is carried out on the basis of neighborhood pixels in the clusters to obtain clustering results corresponding to the pixels; and combining the clustering results corresponding to the pixels to obtain an obstacle detection result.

Optionally, the obtaining module 901 is specifically configured to obtain a plurality of images around the vehicle body;

the clustering module 905 is specifically configured to align grid maps corresponding to multiple times respectively to obtain multiple aligned pixels; aiming at each aligned pixel, fusing the occupation probabilities of the aligned pixels in different grid maps to obtain the occupation probability after fusion; and clustering the aligned pixels based on the fused occupation probability to obtain an obstacle detection result.

Optionally, the first determining module 902 is specifically configured to input the image around the vehicle body into a preset detection network, and output the ground point fractions of the points in the image around the vehicle body through the preset detection network, where the ground point fractions represent the fractions of points that are the ground points of the obstacle; determining an obstacle grounding point in the image around the vehicle body based on the grounding point fraction.

Optionally, the conversion module 903 is specifically configured to project the grounding points of the obstacles in the camera coordinate system to the spherical coordinate system to obtain grounding points of the obstacles in the spherical coordinate system; calculating the coordinates of the grounding points of the obstacles in the spherical coordinate system in the ground coordinate system by utilizing the geometric similarity relation; and performing coordinate conversion on coordinates of the grounding points of the obstacles in the ground coordinate system to obtain the positions of the grounding points of the obstacles in the grid map coordinate system.

Optionally, the obtaining module 901 is specifically configured to obtain an image around a vehicle body collected by a fisheye camera;

the conversion module 903 is specifically configured to project the grounding point of the obstacle in the coordinate system of the fisheye camera to the spherical coordinate system, so as to obtain the grounding point of each obstacle in the spherical coordinate system.

In the technical scheme of the disclosure, the collection, storage, use, processing, transmission, provision, disclosure and other processing of the personal information of the related user are all in accordance with the regulations of related laws and regulations and do not violate the good customs of the public order.

The present disclosure also provides an electronic device, a readable storage medium, and a computer program product according to embodiments of the present disclosure.

FIG. 10 illustrates a schematic block diagram of an example electronic device 1000 that can be used to implement embodiments of the present disclosure. Electronic devices are intended to represent various forms of digital computers, such as laptops, desktops, workstations, personal digital assistants, servers, blade servers, mainframes, and other appropriate computers. Electronic devices may also represent various forms of mobile devices, such as personal digital processors, cellular telephones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions, are meant to be examples only, and are not meant to limit implementations of the disclosure described and/or claimed herein.

As shown in fig. 10, the apparatus 1000 includes a computing unit 1001 that can perform various appropriate actions and processes according to a computer program stored in a Read Only Memory (ROM) 1002 or a computer program loaded from a storage unit 1008 into a Random Access Memory (RAM) 1003. In the RAM 1003, various programs and data necessary for the operation of the device 1000 can also be stored. The calculation unit 1001, the ROM 1002, and the RAM 1003 are connected to each other by a bus 1004. An input/output (I/O) interface 1005 is also connected to bus 1004.

A number of components in device 1000 are connected to I/O interface 1005, including: an input unit 1006 such as a keyboard, a mouse, and the like; an output unit 1007 such as various types of displays, speakers, and the like; a storage unit 1008 such as a magnetic disk, an optical disk, or the like; and a communication unit 1009 such as a network card, a modem, a wireless communication transceiver, or the like. The communication unit 1009 allows the device 1000 to exchange information/data with other devices through a computer network such as the internet and/or various telecommunication networks.

Computing unit 1001 may be a variety of general and/or special purpose processing components with processing and computing capabilities. Some examples of the computing unit 1001 include, but are not limited to, a Central Processing Unit (CPU), a Graphics Processing Unit (GPU), various dedicated Artificial Intelligence (AI) computing chips, various computing units running machine learning model algorithms, a Digital Signal Processor (DSP), and any suitable processor, controller, microcontroller, and so forth. The calculation unit 1001 executes the respective methods and processes described above, such as the obstacle detection method. For example, in some embodiments, the obstacle detection method may be implemented as a computer software program tangibly embodied in a machine-readable medium, such as the storage unit 1008. In some embodiments, part or all of a computer program may be loaded onto and/or installed onto device 1000 via ROM 1002 and/or communications unit 1009. When the computer program is loaded into the RAM 1003 and executed by the computing unit 1001, one or more steps of the obstacle detection method described above may be performed. Alternatively, in other embodiments, the computing unit 1001 may be configured to perform the obstacle detection method by any other suitable means (e.g., by means of firmware).

Various implementations of the systems and techniques described here above may be implemented in digital electronic circuitry, integrated circuitry, field Programmable Gate Arrays (FPGAs), application Specific Integrated Circuits (ASICs), application Specific Standard Products (ASSPs), system on a chip (SOCs), complex Programmable Logic Devices (CPLDs), computer hardware, firmware, software, and/or combinations thereof. These various embodiments may include: implemented in one or more computer programs that are executable and/or interpretable on a programmable system including at least one programmable processor, which may be special or general purpose, receiving data and instructions from, and transmitting data and instructions to, a storage system, at least one input device, and at least one output device.

Program code for implementing the methods of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general purpose computer, special purpose computer, or other programmable data processing apparatus, such that the program codes, when executed by the processor or controller, cause the functions/operations specified in the flowchart and/or block diagram to be performed. The program code may execute entirely on the machine or partially on the machine, partially on the machine and partially on the remote machine or entirely on the remote machine or server as a stand-alone software package.

In the context of this disclosure, a machine-readable medium may be a tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device. The machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of a machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM) erasable programmable read-only memory (EPROM or flash memory), optical fiber, compact disc read-only memory (CD-ROM), optical storage, magnetic storage, or any suitable combination of the foregoing.

To provide for interaction with a user, the systems and techniques described here can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to a user; and a keyboard and a pointing device (e.g., a mouse or a trackball) by which a user can provide input to the computer. Other kinds of devices may also be used to provide for interaction with a user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form, including acoustic, speech, or tactile input.

The systems and techniques described here can be implemented in a computing system that includes a back-end component (e.g., as a data server), or that includes a middleware component (e.g., an application server), or that includes a front-end component (e.g., a user computer having a graphical user interface or a web browser through which a user can interact with an implementation of the systems and techniques described here), or any combination of such back-end, middleware, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: local Area Networks (LANs), wide Area Networks (WANs), and the Internet.

The computer system may include clients and servers. A client and server are generally remote from each other and typically interact through a communication network. The relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other. The server may be a cloud server, a server of a distributed system, or a server with a combined blockchain.

The disclosed embodiment also provides a vehicle, including: such as the electronic device shown in the embodiment of fig. 10.

The vehicle may include an autonomous automobile or the like.

The embodiment of the present disclosure further provides a cloud control platform, including: such as the electronic device shown in the embodiment of fig. 10.

The cloud control platform executes processing at the cloud end, and electronic equipment included in the cloud control platform can acquire data of the vehicle, such as pictures, videos and the like, so that image video processing and data calculation are performed; the cloud control platform can also be called a vehicle-road cooperative management platform, an edge computing platform cloud computing platform, central system, cloud server, etc.

It should be understood that various forms of the flows shown above, reordering, adding or deleting steps, may be used. For example, the steps described in the present disclosure may be executed in parallel or sequentially or in different orders, and are not limited herein as long as the desired results of the technical solutions disclosed in the present disclosure can be achieved.

The above detailed description should not be construed as limiting the scope of the disclosure. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions may be made in accordance with design requirements and other factors. Any modification, equivalent replacement, and improvement made within the spirit and principle of the present disclosure should be included in the scope of protection of the present disclosure.

Claims

1. An obstacle detection method comprising:

acquiring an image around a vehicle body;

determining an obstacle grounding point in the image around the vehicle body;

determining an occupation probability of each pixel in a grid map based on the position of the grounding point of each obstacle, wherein the occupation probability represents the probability that the pixel is occupied by the obstacle, and one grid in the grid map corresponds to one pixel;

2. The method of claim 1, wherein the determining an occupancy probability for each pixel in the grid map based on the location of each obstacle ground point comprises:

aiming at the position of a grounding point of each obstacle, connecting the origin of a grid map coordinate system with the position to obtain a grounding point connecting line;

selecting a pixel area, of which the angle between the pixel area and the grounding point connecting line is smaller than a preset angle, in the grid map;

determining pixels in the pixel area, wherein the distance between the pixels in the pixel area and the position is smaller than a preset distance, and taking the pixels in the pixel area, wherein the distance between the pixels in the pixel area and the position is smaller than the preset distance, as pixels to be updated;

increasing the occupation probability of the pixel to be updated.

3. The method of claim 1, wherein the clustering the plurality of pixels based on the probability of occupancy for each pixel in the grid map to obtain the obstacle detection result comprises:

for each pixel, in response to the fact that the occupation probability of the pixel is larger than a preset occupation probability threshold value, traversing the neighborhood of the pixel, and storing the neighborhood pixels of which the occupation probability is larger than the preset occupation probability threshold value in the neighborhood into the corresponding clusters of the pixel;

clustering is carried out on the basis of neighborhood pixels in the clusters to obtain a clustering result corresponding to the pixels;

combining the clustering results corresponding to the pixels, and obtaining the obstacle detection result.

4. The method of claim 1, wherein said acquiring the vehicle body surroundings image comprises:

acquiring a plurality of images around a vehicle body;

the clustering of the plurality of pixels based on the occupation probability of each pixel in the grid map to obtain the obstacle detection result includes:

aligning the grid maps respectively corresponding to a plurality of moments to obtain a plurality of aligned pixels;

aiming at each aligned pixel, fusing the occupation probabilities of the aligned pixels in different grid maps to obtain the occupation probability after fusion;

and clustering a plurality of aligned pixels based on the fused occupation probability to obtain an obstacle detection result.

5. The method of claim 1, wherein the determining an obstacle grounding point in the image around the vehicle body comprises:

inputting the image around the vehicle body into a preset detection network, and outputting grounding point fractions of all points in the image around the vehicle body through the preset detection network, wherein the grounding point fractions represent fractions of points which are grounding points of the obstacle;

determining an obstacle grounding point in the vehicle body surroundings image based on the grounding point fraction.

6. The method of any one of claims 1 to 5, wherein the coordinate transformation of the obstacle grounding points in the image coordinate system to obtain the position of each obstacle grounding point in the grid map coordinate system comprises:

projecting the grounding points of the obstacles in the camera coordinate system to a spherical coordinate system to obtain the grounding points of the obstacles in the spherical coordinate system;

calculating coordinates of grounding points of all obstacles in the spherical coordinate system in a ground coordinate system by using the geometric similarity relation;

and performing coordinate conversion on the coordinates of the grounding points of the obstacles in the ground coordinate system to obtain the positions of the grounding points of the obstacles in the grid map coordinate system.

7. The method of claim 6, wherein said acquiring the vehicle body surroundings image comprises:

acquiring an image around the vehicle body acquired by a fisheye camera;

8. An obstacle detection device comprising:

the acquisition module is used for acquiring images around the vehicle body;

9. The apparatus according to claim 8, wherein the second determining module is specifically configured to connect an origin of a grid map coordinate system with the position for a position of a grounding point of each obstacle, to obtain a grounding point connection line; selecting a pixel area, of which the angle between the pixel area and the grounding point connecting line is smaller than a preset angle, in the grid map; determining pixels in the pixel area, wherein the distance between the pixels in the pixel area and the position is smaller than a preset distance, and taking the pixels in the pixel area, wherein the distance between the pixels in the pixel area and the position is smaller than the preset distance, as pixels to be updated; increasing the occupation probability of the pixel to be updated.

10. The apparatus according to claim 8, wherein the clustering module is specifically configured to, for each pixel, traverse a neighborhood of the pixel in response to the probability of occupation of the pixel being greater than a preset probability of occupation threshold, and store neighborhood pixels in the neighborhood whose probability of occupation is greater than the preset probability of occupation threshold into the cluster corresponding to the pixel; clustering is carried out on the basis of neighborhood pixels in the clusters to obtain clustering results corresponding to the pixels; and combining the clustering results corresponding to the pixels to obtain an obstacle detection result.

11. The apparatus according to claim 8, wherein the acquisition module is in particular adapted to acquire a plurality of images of the surroundings of the vehicle body;

the clustering module is specifically configured to align grid maps corresponding to a plurality of moments respectively to obtain a plurality of aligned pixels; aiming at each aligned pixel, fusing the occupation probabilities of the aligned pixels in different grid maps to obtain the occupation probability after fusion; and clustering a plurality of aligned pixels based on the fused occupation probability to obtain an obstacle detection result.

12. The apparatus of claim 8, wherein the first determining module is specifically configured to input the image around the vehicle body into a preset detection network, and output a grounding point fraction of each point in the image around the vehicle body through the preset detection network, the grounding point fraction representing a fraction that the point is a grounding point of an obstacle; determining an obstacle grounding point in the image around the vehicle body based on the grounding point fraction.

13. The apparatus according to any one of claims 8 to 12, wherein the conversion module is specifically configured to project the grounding points of the obstacles in the camera coordinate system to the spherical coordinate system, so as to obtain the grounding points of the obstacles in the spherical coordinate system; calculating the coordinates of the grounding points of the obstacles in the spherical coordinate system in the ground coordinate system by utilizing the geometric similarity relation; and performing coordinate conversion on coordinates of the grounding points of the obstacles in the ground coordinate system to obtain the positions of the grounding points of the obstacles in the grid map coordinate system.

14. The device according to claim 13, wherein the acquisition module is specifically configured to acquire an image of a vehicle body periphery acquired by a fisheye camera;

the conversion module is specifically used for projecting the grounding points of the obstacles in the coordinate system of the fisheye camera to the spherical coordinate system to obtain the grounding points of the obstacles in the spherical coordinate system.

15. An electronic device, comprising:

at least one processor; and

a memory communicatively coupled to the at least one processor; wherein,

the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-7.

16. A non-transitory computer readable storage medium having stored thereon computer instructions for causing the computer to perform the method of any one of claims 1-7.

17. A computer program product comprising a computer program which, when executed by a processor, implements the method according to any one of claims 1-7.

18. A vehicle, comprising: the electronic device of claim 15.

19. A cloud controlled platform comprising the electronic device of claim 15.